1. Field of the Invention
The present invention generally relates to semiconductor integrated circuits and, more particularly to integrated circuits having a three-dimensional (3-D) structural configuration including a plurality of semiconductor circuit layers.
2. Description of the Prior Art
It has long been recognized that increased proximity between elements and circuits in electronic devices has numerous benefits in terms of cost, performance and functionality, particularly in complex digital processing circuits integrated on one or more semiconductor chips. Increased proximity of elements allows more devices and circuits to be formed on a single chip using a given process sequence, reducing cost of manufacture per circuit element while increasing functionality of a single chip. Increased proximity of elements on a chip and increased proximity of chips reduces connection length and capacitance and thus improves noise immunity, reduces signal propagation time and improves fan-out (the number of inputs to other circuits that a given circuit can drive). Further, proximity of elements on a chip can be increased as the size of individual elements is reduced, allowing more circuits having more diverse functions to be placed on a chip of given area.
However, the types of circuits which can be included on a single chip may be limited by the technologies required for different types of circuits. For example, the technologies preferred for logic circuits and high density memory circuits are sufficiently different that significant compromises must generally be made in regard to one or both types of circuits in order to fabricate both types of circuit on a single chip.
The combination of these advantages has driven the design of integrated circuit chips to increased integration density and the design of integrated circuit packaging to designs providing increased proximity of chips where a plurality of chips are included in a single package (which provides the added advantage of reducing cost of installing chip packages in various electronic devices since the number of chip packages is reduced for a given device by including a plurality of chips in a single package). However, each input/output (I/O) connection on each chip must include circuitry capable of driving or receiving an off-chip input to another chip. Off-chip signals must be significantly different from on-chip signals in order to maintain noise immunity and freedom from corruption of signals coupled from neighboring connections of relatively greater average length.
One of the more successful and potentially advantageous multi-chip package configurations has been the so-called chip stack where a plurality of individual chips are stacked, connections made directly between the chips and the chip stack packaged as a unit. In such a configuration, connection length between chips is minimized and many circuit topologies are made possible which are not conveniently available on a single, planar (e.g. 2-D) chip, even using complex multi-layer metallization layers to form connections on the chip. Further, chips fabricated using different technologies which would be incompatible on a single chip may be stacked and packaged together as a single unit.
Regardless of the theoretical potential advantages of such chip stack configurations, several highly intractable and interrelated problems have been encountered in practice which have grown even more intractable as reduced minimum feature size regimes and increased integration density for individual chips have developed to support extreme clock cycle speeds. For example, increased chip complexity generally implies an increased number of I/O connections to and from the chip which, in turn, implies a reduced size and increased proximity of I/O connections. Reduced size of I/O connections has several implications such as difficulty of making reliable connections between stacked chips or placement of probes for testing at the wafer or chip level (since the registration tolerance of the testing probe with the chip must be less than the connection pad size/pitch while the connection pad size and spacing must be severely limited to accommodate the number of connections required by increased chip complexity and while increased proximity of I/O connections increases signal coupling).
Moreover, increased integration density and reduction of circuit element size requires more stringent protection of internal chip circuits from overvoltage, electrostatic discharge (ESD) and the like at points where I/O connections are made. Additionally, broadband impedance matching circuits such as a T-coil device requiring a planar inductor may be needed and which may require substantial chip area, especially when replicated for each I/O connection. That is, the extra load presented by the introduction of ESD structures makes it more likely that additional impedance matching circuits may be needed to assure proper electrical behavior at the I/O and such impedance matching circuits typically include a planar inductor that may require substantial area. Such protection and impedance matching circuits must be provided in addition to circuits capable of providing signals suitable for off-chip device drive and, with increasing likelihood, signal level conversion, clock signal conversion and buffering circuits, consuming further chip area.
At the same time, increased chip complexity increases the need for testing at the chip level since a chip stack will be defective if any single chip included therein contains a defect while increased proximity of connection pads coupled with the fact that a chip stack remains a packaging configuration for individual chips and I/O circuits capable of driving off-chip circuits must be provided for each connection pad increases the likelihood of signal coupling between neighboring connections to each chip. Further, as increased integration density has allowed higher clock cycle rates reaching into the Gigahertz range, signal capture and analysis using external testers has become much more difficult, partly due to the size thereof, while more extensive tests are required by increased chip complexity that are correspondingly less well accommodated by test arrangements that can be provided in stringently limited area on the chip itself. Moreover, the extreme clock cycle speeds now possible require routing of signals from outputs of the chip to inputs of the chip over very short but wide (e.g. four bytes or more) bus or with buffering or latching arrangements (e.g. to limit effects of signal propagation time on the test results) which interfere with core logic layout. Thus the provision of more numerous circuits capable of providing protection, off-chip level drive and various needed signal conversions as well as routing has become increasingly expensive and impractical while testing has become much more expensive and impractical even in the face of increased need therefor.
In this latter regard, in addition to the practical problem of probing more densely integrated circuits having necessarily reduced connection pad dimensions alluded to above, the capture and isolation of a signal of interest in the Gigahertz range using an external tester presents such extreme difficulties that an external tester capable of doing so may cost several hundreds of thousands of dollars per I/O connection or pin. At the same time, testing at such frequencies has implications on chip design and layout and proximity of related input and output connections (e.g. providing an on-chip wrap bus to wrap the outputs to inputs which may interfere with core logic functions) which are difficult to meet and which engender other complications of signal capture and isolation, particularly at off-chip drive signal levels.
In addition to testing of chips during integrated circuit manufacture, it is desirable to periodically test a chip for functionality after the chip is placed in service (e.g. at power-up of the chip). For this purpose, so-called built-in self-test (BIST) arrangements have been developed and at the present state of the art are quite sophisticated. However, BIST arrangements, by their nature, impose a trade-off between the extent of testing that can be done, the amount of time which can be allotted to a self-test procedure and the amount of chip space that can be allocated to the BIST arrangement, particularly since the BIST arrangement must provide for generation (e.g. generally from storage) for the bit patterns which will be used to exercise the circuit and the bit patterns against which the result from a test bit pattern is to be compared to confirm functionality. Therefore, the chip area allocated to the BIST arrangement is generally limited to about two percent of the chip area and the extent of testing available is usually severely limited and reliance of external testers is generally required to confirm full functionality of a chip. Furthermore, BIST arrangements are used only in the digital part of the integrated circuit chip, which might also include analog functions as well such as receivers or transmitters or mixed-signal digital/analog functions, such as converters, for which different isolation strategies and testing structures are needed.
Therefore, particularly in view of the need for larger and more complex protection circuits at I/O connections and the need for providing testing at the chip or wafer level in order to avoid the much increased costs of later-discovered defects in individual chips after chip stacks have been formed, there has been no practical alternative to simply omitting chip and wafer level testing altogether and accepting the increased costs of rejecting substantially completed multi-chip packages during testing at the package or system level. Moreover, known techniques of chip stacking cannot, at practical costs, support current and foreseeable chip technologies and integration densities while already severely compromising the benefits which could potentially be derived from currently possible technologies and integration densities due to the cost of providing suitable I/O protection and off-chip drive signal levels at I/O connections between chips, even when commonly packaged.
This latter problem is particularly aggravated in regard to application specific integrated circuits (ASICs) which, as a practical matter, are generally designed using so-called cores or macros which are pre-designed functional circuit portions which can be laid out and connected on a single chip to provide, in the aggregate, a customized function which must often include circuits and connections for converting standard bus designs of existing cores to a customized bus configuration, often provided as a separate so-called bridge chip. As ASIC designs become more complex and higher performance levels are required, it would be desirable, particularly from the standpoints of manufacturing cost and potential performance improvement, to utilize a chip stack configuration for ASICs with individual cores or small groups of cores on separate chips. Such a configuration would enable better design re-use, faster time to market and quick customization of a design as well as the performance benefits alluded to above. However, as noted above, each chip must then include I/O connections including circuit protection and circuits capable of providing off-chip level drive signals. In practical effect, such a provision would amount to providing interfaces between virtually all cores in addition to the cores themselves (and any bus conversion chip included in the ASIC) and would be of prohibitive cost.
In summary, while it is seen that the above problems with common packaging of multiple chips, and chip stacks in particular, derive from the fact that each chip must be regarded as an off-chip device relative to every other chip and a suitable inter-chip interface provided while the requirements for each connection of each such interface have become more complex and stringent and consume increased chip area for each of an increased number of connections to the point of becoming prohibitive. To solve this basic problem, a so-called three-dimensional chip has been developed including a vertical bus arrangement such that drive signals suitable for off-chip devices are not needed and the separately fabricated chips effectively and functionally become portions of the same chip folded over on itself with the physical configuration providing the same potential advantages of chip stack configurations. However, such a connection does not solve the more specific and intractable problems of bus design conversion, customization and translation, the provision of suitable protection and off-chip device drive or, especially, chip or wafer level testing to support current and foreseeable chip design and manufacturing technology and the advantages potentially available therefrom.